You can deploy this model and start earning money today!

WhiteRoomProdigy/
amicus-humanizer-v1

Name: amicus-humanizer-v1
Rating: 2.5 (16 reviews)

safetensors

text2text-generation

text-generation-inference

custom-legal-human-corpus

apache-2.0

Views

No views yet

🏛️ Amicus Humanizer v1 (8-bit)

Amicus Humanizer v1 is a specialized text2text-generation model designed to rewrite AI-generated text to sound entirely human. Built on the highly efficient T5 architecture and quantized to 8-bit precision using bitsandbytes, this model is explicitly fine-tuned to bypass AI detectors while strictly preserving domain-specific meaning, legal citations, and professional tone.

Developed by the team at Dockase, this open-source release aims to empower researchers, law firms, and legal tech developers to seamlessly convert robotic, predictable AI drafts into authoritative, human-sounding prose.

🚀 Key Features

AI Detection Bypass: Restructures syntax, injects "burstiness" (sentence variety), and eliminates predictable AI tropes to seamlessly bypass detectors like GPTZero, Turnitin, and Copyleaks.
Domain-Specific Preservation: Unlike generic paraphrasers, Amicus is trained to never alter strict legal terminology, case citations, or core factual arguments.
8-Bit Quantization: Shipped in 8-bit precision (bitsandbytes), reducing VRAM requirements drastically. It runs lightning-fast on cheap consumer GPUs (like the NVIDIA T4) with zero degradation in output quality.
Active Voice Prioritization: Automatically converts passive, bloated AI phrasing ("It is crucial to remember that...") into sharp, authoritative active voice.

🛠️ Usage & Inference

Because the model is quantized using bitsandbytes, you can load it instantly with Hugging Face transformers.

Installation

pip install transformers accelerate bitsandbytes

Python Example

python

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

model_id = "WhiteRoomProdigy/amicus-humanizer-v1"

# Load the tokenizer and 8-bit model
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSeq2SeqLM.from_pretrained(model_id, load_in_8bit=True, device_map="auto")

# Input text (e.g., rigid AI-generated legal text)
input_text = "Furthermore, it is important to note that the plaintiff failed to establish a breach of contract."

# Tokenize and Generate
inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_length=512, do_sample=True, temperature=0.7)

# Decode the humanized output
humanized_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(humanized_text)

🧠 Model Architecture

Base Model: T5 (Text-to-Text Transfer Transformer)
Task: Text-to-Text Generation (Paraphrasing / Humanization)
Quantization: 8-bit (bitsandbytes)
Language: English
License: Apache 2.0

⚖️ Intended Use & Limitations

Intended Use:

Rewriting first-draft AI legal memos into finalized, natural-sounding documents.
Restructuring rigid prose to improve readability and flow.
Bypassing false-positive AI detection on legitimate professional writing.

Limitations:

While it strongly preserves meaning, outputs should always be reviewed by a human expert before use in official legal filings.
The model is primarily trained on English legal and professional text; performance may drop on highly informal slang or non-English languages.

🏢 About Dockase

This model was developed by Dockase, the fundamental legal operating system and workflow automation stack for the African market. We are committed to open-sourcing powerful, domain-specific AI tools to push the boundaries of legal tech.